Skip to content

Latest commit

 

History

History
83 lines (70 loc) · 4.37 KB

integration-testing.md

File metadata and controls

83 lines (70 loc) · 4.37 KB

Public integration testing

StatLite's documented integrations are continuously exercised in public CI with real StatLite collections. Most cases start the shipped runnable example for that guide. The Caddy case starts pinned Caddy 2.11.4 and two temporary Python servers that match the documented named-site shape.

How it works

Each guide is the canonical integration recipe. For the example-backed guides, the public workflow starts that example, so users can inspect and run the fixture used by the check.

Each integration case starts the application and StatLite, generates bounded traffic, and forces one or more collections. The exact ordering varies by adapter. In general, the workflow covers:

  • Building or installing the example and waiting for its endpoint to be ready.
  • Starting the current StatLite build with the example configuration and isolated temporary storage.
  • Sending a small deterministic set of normal and error requests.
  • Forcing one or more collections through StatLite's existing debug API instead of waiting for the polling interval.
  • Checking the important basic request, error, and latency metrics, then confirming that results are visible through StatLite's summary and series APIs.

The direct statlite-metrics/v1 cases also run statlite inspect, verify the application's metrics endpoint, and confirm that polling that endpoint does not increase application counters. The Go net/http and Gin cases establish a baseline, then assert exact two-poll counter deltas and derived average latency, including Gin's recovered pre-commit panic. The Spring Boot and Quarkus cases check the framework-specific health or process/runtime signals that their examples publish. The Micronaut case checks typed inspection, exact 404/4xx/5xx deltas between successful stored polls, accumulated duration, runtime signals, aggregate UP health, and unavailable database health. Management self-traffic contributes to its total, so the total increase is a lower bound. The Caddy case first checks that collection fails with caddy_http_not_observed before either site has traffic, then checks exact request, 404, 4xx, and 5xx deltas for two hosts that share one admin metrics endpoint. Each host stays isolated, and each request is served by its own upstream.

Integrations in the public checks

The shared matrix currently exercises:

Inspect the implementation and current triggers in the public integration certification workflow. It runs for every pull request, every push to main, and weekly on Monday at 09:17 UTC. Matrix cases use fail-fast: false, so one integration failure does not hide the results for the others.

The release workflow reuses this same workflow as a prerequisite before it creates a release tag, so a release cannot proceed until all nine public cases pass on the dispatched commit.

These are focused public integration checks for the documented monitoring journey. They do not cover every framework version, production topology, deployment environment, or application behavior, and they are not a compatibility guarantee. Broader private and release testing remains separate where additional coverage is needed.